The A.I. boom bears every hallmark of a bubble. Hyperscalers are spending hundreds of billions on data centres, electricity demand is spiking at 40 per cent annual growth rates, and start-ups with uncertain business models are commanding valuations usually reserved for oil giants. These are not signs of sober allocation of capital — they are the familiar contours of speculative excess. Yet the critical question is not whether this is a bubble, but what follows when it inevitably bursts.

History suggests the answer is less apocalyptic than investors fear.

Lessons from 2000

The most obvious parallel is the dot-com crash. In March 2000, the Nasdaq began a collapse that wiped out $5 trillion in market value. Household names such as Pets.com disappeared. Yet the underlying technology — the internet — not only survived but went on to define the global economy. Amazon, Microsoft and Apple each lost more than 65 per cent of their value at the time, with some taking over a decade to recover, but they emerged as the most powerful companies of the 21st century.

The key lesson: bubbles destroy capital, not innovation. The excess spending of the late 1990s laid the groundwork for the fibre-optic networks and infrastructure that powered the internet economy of the 2010s. What initially appeared to be waste became the foundation of trillion-dollar markets.

The key lesson: bubbles destroy capital, not innovation

Today’s Overbuild, Tomorrow’s Advantage

The same logic applies to A.I.. The capital intensity of today’s boom is striking. Goldman Sachs projects that U.S. hyperscalers will invest $736 billion in 2025 and 2026 alone. Bank of America estimates A.I.-related electricity demand will expand more than 40 per cent annually. Such outlays are unsustainable, and investors will eventually baulk at low conversion rates and unproven monetisation models.

But the infrastructure will remain. When the shakeout comes, data centres, semiconductor fabs, and high-capacity grids will already be in place, assets that competitors in Europe and Asia will struggle to match. Just as overinvestment in shale oil created years of cheap U.S. energy, today’s A.I. arms race may lay the groundwork for America’s technological dominance.

The infrastructure will remain

Equity Bubbles vs Debt Bubbles

Another reason for restraint rather than panic: this is an equity-driven bubble, not a debt-fuelled one. The dot-com crash wiped out investors but left banks intact, limiting the damage to the broader economy. By contrast, the 2008 financial crisis was a debt bubble — its collapse paralysed credit markets and triggered a systemic recession.

The A.I. boom is closer to 2000 than 2008. Pain will be concentrated among shareholders and venture funds, while the broader economy benefits from lower computing costs and cheaper access to machine learning infrastructure.

The New Power Equation

Bubbles are not just financial phenomena; they reshape power structures. The railway bubble of the 1840s bankrupted many investors but left Britain with the transport backbone of the industrial age. The telecom bubble of the 1990s made internet access ubiquitous, even as operators wrote off billions.

The coming A.I. correction will follow the same pattern. Some of today’s high-fliers will vanish, but those that endure — Microsoft, Google, Nvidia, Amazon — will emerge even more entrenched. Meanwhile, new entrants born during the downturn will inherit infrastructure they could never have afforded to build. The result will be a second wave of innovation, likely more sustainable than the first.

What to Expect

When the A.I. bubble bursts, valuations will contract, capital expenditure will slow, and a generation of speculative start-ups will vanish. That is not a crisis; it is the normal arc of technological revolutions. Investors will lick their wounds, policymakers will claim vindication, and entrepreneurs will quietly get back to work.

That is not a crisis; it is the normal arc of technological revolutions

The internet crash did not prevent the rise of e-commerce. The railway bust did not halt industrialisation. The A.I. correction will not derail machine learning’s integration into business, government and daily life. It will simply separate hype from utility, and history suggests that is when the real value creation begins.

Subscribe now

Building Creative Machines

When the AI Bubble Pops, What Happens Next?